Basic mechanism of transcription by RNA polymerase II

Abstract

RNA polymerase II-like enzymes carry out transcription of genomes in Eukaryota, Archaea, and some viruses. They also exhibit fundamental similarity to RNA polymerases from bacteria, chloroplasts, and mitochondria. In this review we take an inventory of recent studiesilluminating different steps of basic transcription mechanism, likely common for most multi-subunit RNA polymerases. Through the amalgamation of structural and computational chemistry data we attempt to highlight the most feasible reaction pathway for the two-metal nucleotidyl transfer mechanism, and to evaluate the way catalysis can be linked to translocation in the mechano-chemical cycle catalyzed by RNA polymerase II.

Introduction

Transcription of cellular genomes in all domains of life is carried out by essentially orthologous enzymes, multi-subunit DNA-dependent RNA polymerases (RNAPs). The archetypal eubacterial core enzyme, represented by that of Escherichia coli, consists of two large subunits (β and β′)^1,2, that carry out the chemical (NTP condensation) and mechanical (translocation) steps^3,4, and three smaller ones, α₂ dimer, and ω, that play roles in the assembly of the enzyme and regulation of transcription^1,2,4,5 (Fig. 1A). Variant forms of the core are described for some Eubacteria, such as Francisella (featuring α₁α₂ heterodimer⁶), and Bacillus (adding δ and alternate ω subunits⁷), whereas in Helicobacter, Wolbachia and Wolinella RNAPs β and β′ subunits are fused together¹. RNAP subunit initially described as γ in Nostoc, Anabaena, and other Cyanobacteria turned out to be a product of a split in the ancestral gene, coding for β′ ortholog^8,9. A similar split exists in the β′ subunit of plastid-encoded RNAPs in chloroplasts, featuring a reduced bacterial-type core α₂ββ′ (ω subunit appears to be encoded by plastid genomes only in Rhodophyta algae)¹⁰. Interestingly, this core architecture predominates in early chloroplast development (e.g. in etioplasts), whereas in mature chloroplasts it is augmented by up to 30 additional subunits¹¹. Mitochondrial RNAPs are predominantly related to those of bacteriophages, but the relatively large mitochondrial genomes of some protists (namely Jakobidae and Malawimonadidae) encode orthologs of bacterial α, β, and β′ subunits^12,13.

Figure 1. Orthologous bacterial RNA polymerase-like core in Eubacteria (A, 3lu0), Eukarya (B, 1nik), and Archaea (C, 3hkz).

All enzymes are surface, light teal, β′-like subunits are deep teal, β-like subunits are yellow-orange, α-like subunits are red and warm pink, ω-like subunits are purple-blue.

Nuclear RNAPs in Eukaryota are represented by a minimal set of three classes: RNAPI transcribing ribosomal RNA genes, RNAPII carrying out the synthesis of messenger RNA and a subset of small non-coding RNAs, and RNAPIII synthesizing transfer RNAs, 5S RNA, and the bulk of small non-coding RNAs^14–17. RNAPs from each class contain 12+ subunits, with a core orthologous to bacterial-type enzyme: yeast RNAPII largest subunits, Rpb1 and Rpb2, correspond to β′ and β, respectively, Rpb3 and Rpb11 are divergent orthologs of α, and Rpb6 is an ω ortholog^18–20 (Fig. 1B). In plants the set of nuclear RNAPs is extended to 5, RNAPIV and V conforming to the overall architecture of RNAPI-III^21–23, but exhibiting an unexpected divergence in the mechano-chemical core^24,25, otherwise universally conserved in both pro- and eukaryotes, which led some to question the competence of these enzymes to act as bona fide RNAPs²⁴ (this competence still eludes in vitro demonstration²²). Genomes of large viruses from the order Megavirales encode RNAPs, related to the nuclear RNAPII, but with a reduced complement of subunits, typically numbering 8²⁶. Remarkably, whereas some enzymes, such as that of Mimivirus, feature a fully conserved α₂ββ′ω-like core, namely the orthologs of Rpb1, 2, 3/11, and 6 (together with Rpb5, 9, and 10 homologs)²⁷, RNAPs from poxviruses lack an apparent α ortholog^28,29, indicating a potentially dramatic change in enzyme architecture and assembly. Lane and Darst, based on the results of their large-scale multiple sequence alignment for β,β′-like subunits, argued that poxvirus RNAPs were related to the nuclear RNAPI enzymes¹. Finally, Archaea contain one type of RNAPs, similar in size (11–13 subunits) and composition to RNAPII: it comprises orthologs of Rpb1 (split into two subunits), 2–8, 10, and 12^30,31 (Fig. 1C). Further in depth analysis of structural and sequence data pertaining to RNAPs from bacteria, archaea, and eukaryotes is provided by Cramer, Darst, Kornberg, Murakami, Yokoyama and colleagues^{1,2,18–20,30–33}. In this review we will focus on the basic mechanism of transcription, as it emerges from studies of yeast RNAPII, complemented when necessary with the relevant data for bacterial enzymes.

Nucleotide addition cycle

Synthesis of RNA by RNAP comprises three distinct stages, initiation, elongation, and termination^33–37. The second and longest stage, elongation, is a reiterative mechano-chemical process that consist of a single NTP condensation reaction, that extends nascent RNA by one NMP with concomitant release of pyrophosphate (PP_i), followed by enzyme translocation along the template by one nucleotide (nt)^36,38,39.

RNAP is different from the classic Cleland enzymes in that it does not dissociate one of the reaction products, RNA, each cycle, it remains bound to the enzyme, and becomes a substrate as the new cycle begins. Nascent RNA is annealed to the template strand of DNA for 8–9 nts, forming so-called RNA-DNA hybrid, which is a major determinant of elongation complex stability^20,40,41. The length of the hybrid is maintained constant throughout the cycle: as the 3′ end extends by one nt, another from the 5′ end of the hybrid peels off the template and is threaded through the RNA exit channel. Read-out of the template, governed by the Watson-Crick complementarity of the incoming NTP and the templating nt, requires melting of the double-stranded DNA to allow formation of the RNA-DNA hybrid. Locally melted DNA segment is known as transcription bubble, with single-stranded DNA chains reaching 11–12 nt in length being engaged by RNAP and the hybrid. Melting of DNA at the front edge of RNAP and its annealing at the back edge are thought to occur simultaneously with translocation and to be neutral in terms of free energy, if averaged over the length of the transcript.

Translocation returns 3′ end of RNA into the catalytically active position (so-called nucleophilic, or i site) and resets the enzyme for accepting the next NTP, so it is imperative that it spans only one nt of template DNA. Hypertranslocation occurs in vitro (stimulated by addition of a cognate NMP), as does retrograde translocation (backtracking), both in vitro and in vivo, indicating that there is no significant activation barrier to either of them^42–46. The mechanism of limiting translocation to one nt step is not known, and so far translocation has not been replicated in molecular dynamics (MD) simulations. Crystallographic models exist, that may have captured translocation intermediates^43,47, but the insights they provided are not sufficient to build an explicit MD or molecular mechanics (MM) trajectory of the translocating RNAP. It has been proposed that the “gating tyrosine” (Rpb2 Tyr769) physically delimits backtracking by sterically hindering base stacking beyond the first backtracked RNA nt⁴². It is also tempting to speculate that deformability of the so-called bridge helix (Rpb1 815–845), long implicated in various aspects of the transcription cycle, determines the single nt step of the forward translocation^47–50. Both models await further experimental and computational interrogations. The size of the transcription bubble also appears to be maintained constant, even during backtracking or hypertranslocation, although it may expand by extending the size of the RNA-DNA hybrid leading to the formation of so called R-loops, detrimental to genome stability in vivo⁵¹.

NTP condensation is believed to occur via a general two-metal-ion mechanism, common for all nucleotidyl transferases, first put forth by Thomas Steitz⁵². The central feature of this mechanism is the employment of two magnesium cations, coordinated by at least two aspartate residues located in the active site. According to this general model the first magnesium (A) promotes deprotonation of the RNA 3′OH, facilitating 3′ O⁻ attack on the substrate NTP α-phosphate, which in turn leads to formation of a new phosphodiester bond and a leaving group, PP_i. The second magnesium ion (B) participates in stabilization of the pentacovalent transition state. This mechanism was further elaborated upon by Goldfarb and colleagues, who proposed the universal central role of two Mg cations coordinated in the RNAP active site in other reactions catalyzed by this enzyme: exo- and endo-ribonuclease, and exo- and endo-pyrophosporylase^53,54. Detailed reaction geometries at the pseudo-atomic resolution were developed, consistent with the bulk biochemical interrogation of the E. coli RNAP. These reactions, although not situated in the main pathway of NTP condensation, play important roles in transcription fidelity, reactivation of arrested complexes, etc^3,55.

Multi-subunit RNAPs feature an invariant triad of aspartate residues^54,56 (Rpb1 Asp481, 483, 485), that coordinates both the persistently bound MgA and the NTP-bound MgB, thought to enter and leave the active site during each NTP addition cycle. Structural analysis of various NTP-bound RNAP complexes extended the boundaries of the active site to include a number of other residues, participating in catalysis, metal coordination, or substrate binding.

Composition of the active site and the details of coordination/interaction network remain a subject of some controversy, stemming from the occasional incongruity of structural models featuring mechanistically identical complexes. The atomistic details inferred from such models must be treated with a caveat in mind, namely that crystallographic models by their nature deal with catalytically inactive mimics of the on-pathway intermediates. In order to prevent catalysis in the NTP-bound RNAP crystals NTPs are often represented by their non-hydrolysable analogs (such as α,β-methylene derivatives GMPCPP or AMPCPP), in other cases the reactive 3′ OH group have been removed from the RNA^57–59. These changes directly affect coordination of the RNA and/or NTP in the active site of the enzyme, whereas these and other interactions are subject to artifacts of crystallization, quality of the electron density and of the subsequent computational structure solution. In one of his recent papers addressing variations in NTP binding mode Patrick Cramer noted that “the exact positioning of the NTP depends on the experimental design”⁵⁸. This remark highlights the fact that crystallographic structural models provide not only many crucial insights into the reaction geometry and dynamics, but also an occasional artifact, singular to a given experimental setup, that can dramatically affect the interpretation of the data. Unfortunately there are limitations on the degree to which such artifacts can be ameliorated through the improvements in the design of crystallographic experiments. So far no crystallographic model of any multi-subunit RNAP features intermediates of the NTP addition reaction, with all available structures representing either pre- or post-chemistry states. An intriguing possibility to explore the reaction mechanism by crystallographic means has been opened up by the successful application of time-resolved X-ray crystallography to the dNTP condensation by DNA polymerase η⁶⁰. Having created crystals poised for catalysis by replacing catalytic Mg cations with inert Ca ones, Nakamura et al. were able to start the dNTP addition by restoring Mg cations to the system and stop it at different times by rapidly cooling the crystals to 77 K. As a result new details of the reaction mechanism emerged, such as previously unobserved and unforeseen third Mg cation in the active site of the enzyme coincident with PP_i formation⁶⁰. Confidence in the assigning electronic density to reactants or catalytic metals is significantly boosted in this case by the high resolution of the crystallographic models, ~ 1.5 Å, whereas the reverse is true for RNAP II, which structures are routinely resolved at 3–4 Å. Hence although NTP condensation by RNAPII should be amenable to similar time-resolved crystallographic analysis, the fine atomistic and mechanistic details of this reaction can only be derived with confidence from the models resolved at < 2 Å. Complementary methods, experimental as well as computational, have to be deployed in order to fully explore the atomic, kinetic and thermodynamic details of the reaction mechanism⁶¹. It bears noting that even the considerable power of modern MD and MM computations cannot mitigate the problems present in the starting structural models, such as poor resolution and/or refinement, which give rise to technically competent but materially spurious simulations. The growing appreciation of the fact that the quality of “structures” (crystallographic models) deposited with the Protein Data Bank varies greatly (but imperceptibly for many of the downstream users) led to the introduction of new and improved structure validation protocols^62,63, whereas comprehensive structure evaluation and rebuilding/refinement are available from such resources as PDBREPORT (http://swift.cmbi.ru.nl/gv/pdbreport/) and PDB_REDO (http://www.cmbi.ru.nl/pdb_redo/)^64,65. These tools can assist with assessment and choice of the crystallographic models, an issue of critical importance as the strength of the structure-based arguments (including computational ones) in many ways depends on the quality of the starting model.

RNA polymerase active site

The minimal composition of the RNAPII active site (Fig. 2) as it emerges from the crystallographic models of the enzyme^{42,43,57,58,66–73} in addition to the catalytic triad (Rpb1 Asp481, 483, 485) includes Rpb2 residues Asp837, that coordinates MgB, Lys987, and Arg1020, which form H-bonds with the substrate triphosphate, Rpb1 His1085 as a candidate general acid^57,74, Arg446, Asn479 and Gln1078 that contact 2′ OH of the substrate and may participate in discrimination of NTPs over dNTPs, and Leu1081 which stacks on the NTP base. Non-peptide elements of the active site consist of 2 catalytic Mg ions, templating nt, that forms Watson-Crick pair with incoming NTP, and 3′ terminal nt in the RNA, which prior to the nucleophilic attack stacks on the base of the substrate NTP^57–59,75. Gln1078, Leu1081, and His1085 are embedded into a metamorphic structural module of the RNAP, the so called trigger loop (TL, Rpb1 residues 1070–1100)^57,76. TL is widely believed to adopt two obligatory conformations: open, with an unstructured/loop tip in absence of NTP, and closed, essentially an α-helical hairpin, which forms upon binding of substrate NTP^{38,57–59,77–79} (Fig. 3). The obligatory transition between two forms of the trigger loop, one of the most deformable/mobile elements of RNAP⁸⁰ is most clearly seen in the bacterial enzyme⁵⁹, whereas in yeast RNAPII it can adopt different conformations even in the presence of the substrate^57,58. Notably, the TL in these complexes tended to adopt the closed conformation in presence of UTP or GTP, and remain in open one when bound to α,β-methylene derivative GMPCPP (Fig. 3), emphasizing the importance of proper phosphate coordination by the trigger loop residues (predominantly His1085)⁷⁹ in binding and catalysis.

Figure 2. Active site of yeast RNA polymerase II (2e2h).

Template DNA is cartoon grey, except for the templating nucleotide (black), RNA is cartoon light pink, except for the 3′ nucleotide (hot pink). Substrate GTP is orange sticks. Catalytic Mg²⁺ cations are solid magenta spheres, catalytic tetrad (Asp481, 483, 485, and 837) is red sticks. Trigger loop is green cartoon, residues Gln1078, Leu1081, and His1085 are green sticks. The rest of amino acids are light teal sticks.

Figure 3. Structural variants of the trigger loop in yeast RNA polymerase II.

Elements of the active site (from GTP-bound 2e2h) are represented as in Fig. 2. (substrate is orange sticks, catalytic Mg²⁺ cations are solid magenta spheres, catalytic tetrad is red sticks) Superimposed are trigger loops (cartoon) from NTP-free enzyme (1sfo, purple blue), GMPCPP-bound 2e2j (yellow) and 2nvt (marine), and UTP-bound 2nvz (red).

Role of the invariant Asn479 in ribo-discrimination, based on structural and biochemical studies of bacterial RNAPs^59,75, was called into question by Kornberg and colleagues based on structural analysis of yeast RNAPII⁵⁷. Recently interactions between 2′ OH and Asn479 were observed in NTP-bound initiation complex of RNAPII⁵⁸ and the apparent discrepancies were explained by the differential NTP coordination due to absence of the 3′ OH in the complexes crystallized by Wang et al.⁵⁷

The Watson-Crick complementarity of the incoming NTP and the templating DNA nt is of paramount and obvious importance for the overall fidelity of transcription. However, the catalytic role of these interactions was not apparent until Kellinger et al. tested a set of non-polar isosteres of the templating nt in the NTP condensation⁸¹. These nucleotide analogs, matching the conventional DNA nts in size (within 1 Å) were completely devoid of the H-bond donors/acceptors, eliminating the need to consider alternate/wobbling bonding patterns. Remarkably, removal of the Watson-Crick complementarity with the incoming NTP reduced the apparent binding constant (K_M) only 6-fold with concomitant drop in the k_cat of about 5600-fold⁸¹. This prominent effect on catalysis clearly indicates that the Watson-Crick complementarity with its planar conformation of the bonding DNA-NTP pair orients the incoming substrate for deprotonation and subsequent nucleophilic attack, arguing for the inclusion of the templating nt among the catalytic elements of the RNAP active site.

Mechanism of NTP condensation

The theory of nucleotidyl transfer via two-metal catalysis does not specify the identity of the proton acceptor for the 3′ OH, nor the same of the proton donor for the PP_i, whereas the proton inventory experiments suggested the general two-proton transfer mechanism for DNA and RNA polymerases and indicated that the PP_i leaves active site in the protonated form^74,82. The available structures of NTP-bound yeast and bacterial RNAPs^{57–59,69,73,83} allow only the exclusion of certain residues from direct participation in catalysis (e.g. as being too far removed from the active site), but not the unambiguous elucidation of reaction mechanism and intermediates. Some of these structures have been recently used in MD and other computational interrogations of the basic mechanism of NTP condensation.

Using the 4.3 Å resolution crystal structure of RNAPII elongation complex (2nvz⁵⁷) as the starting model for MD simulations Zhu and Salahub explored some of the possible mechanisms of 3′ OH deprotonation⁸⁴. To this end they employed a first-principles-based reactive force field, ReaxFF, parameterized for proteins and nucleic acids, and a backbone model, including incoming NTP, 3′-terminal nt of the RNA, catalytic loop (Rpb1 481–485), His1085, and two Arg residues (766 and 1020). This model was employed in simulation of the general base variation on the nucleophilic attack, where the 3′ OH attacks the α-phosphate of the incoming NTP and is simultaneously deprotonated by the action of the catalytic residue, Asp485. The activation barrier for this reaction pathway was estimated at 19 kcal/mol.

In an alternate simulation (involving only the catalytic loop, and the 3′ terminal nt of RNA) 3′ OH was deprotonated by the bulk solvent (water) generating nucleophile prior to the attack. Activation barrier to this step was only 7 kcal/mol. Further emphasizing the feasibility of this step was the observation of water molecules clustering around the 3′ OH in the equilibrium simulation⁸⁴. Subsequent nucleophilic attack by the 3′ O⁻ on the NTP was estimated at 17 kcal/mol, thus lowering the highest activation barrier by 2 kcal/mol. Notably, regardless of the deprotonation pathway, the dissociation of the PP_i was determined to be the rate-limiting step, with activation energy of 23–24 kcal/mol⁸⁴. These calculations did not take into account the exit of PP_i in the protonated form, nor the potential assisting role played by the positively charged residues in the secondary channel (the likely exit route for PP_i), both of which can dramatically alter the numerical values of the activation barrier. The models used in these simulations were also quite abbreviated, leaving out some catalytically important residues (e.g. Asp837). This would not impact the results of the water-assisted deprotonation of the 3′ OH as much as the simulations of the nucleophilic attack and the PP_i dissociation.

In their later work Salahub and colleagues supplemented ReaxFF computations with density functional theory (DFT) simulations in analysis of the RNAPII active site, focusing on the catalytic loop (DFDGD motif) and the Mg²⁺ coordination⁸⁵. They have used 3.3 Å structure of the pre-translocated RNAPII elongation complex (EC) (1i6h⁶⁷) to build models of the catalytic loop in the RNA-free RNAPII, bound to DNA, and the pre-translocated EC; the model of the catalytic loop in the post-translocated form was built using 4.3 Å structure of the UTP-bound EC (1r9s⁶⁹). DFT simulations were limited to 57, 91, and 108 atom models, whereas ReaxFF calculations were based on 250 atom model of the post-translocated EC, developed previously⁸⁴, solvated with 400 water molecules⁸⁵. Through the combination of these approaches Salahub et al. were able to generate an improved equilibrium model of Mg²⁺ and NTP coordination in the active site of RNAPII in solution⁸⁵.

Comparing pre- and post-translocated models, the authors noted structural expansion-contraction of the catalytic loop in the EC depending on the presence of the substrate NTP: it becomes more compact upon binding of NTP and expands upon completion of the chemical step, all the while stabilized by an ordered network of H-bonds, connecting carboxylate oxygen of one Asp residue to the backbone amide of the neighboring one⁸⁵. Notably, a contraction of the catalytic loop was reported by Vassylyev and colleagues, who compared the structures of the apo-holoenzyme RNAP from Thermus thermophilus and the holoenzyme bound by the antibiotics rifabutin and rifapentin; they have linked this contraction to the apparent concomitant loss of catalytic Mg²⁺ from the active site⁸⁶ (this interpretation was later challenged by Ebright and colleagues⁸⁷). Changes in Mg²⁺ coordination due to the binding of transcription factors GreA/B (bacteria) and TFIIS (eukaryotes) and introduction of new potential Mg²⁺-coordination partners (both Gre and TFIIS feature conserved acidic residues they deposit into the active site upon binding to RNAP) have been proposed as the mechanism of the “remodeling” of the active site from RNA synthesis to transcript cleavage^72,88,89. Structurally similar to Gre cleavage factors, the inhibitory protein Gfh1, was proposed by Darst, Vassylyev and colleagues to coordinate Mg²⁺ in a catalytically inactive configuration^90,91 (for alternate mechanism of Gfh1 action see Tagami et al.⁴⁷). The central role Mg ions play in all these scenarios, the proposed changes in efficiency and even the direction of the catalyzed reaction potentially caused even by a subtle alteration in Mg²⁺ coordination^92–94 necessitate rigorous exploration of reaction mechanisms involving these cations.

Lately the universality of the two-metal catalysis scheme, that owes its inception and broad acceptance to structural work, has been challenged based on biochemical data and computational simulations^95,96, and the presence of His residue (His1085 in yeast RNAPII) in the catalytic site apparently makes RNAP a subject for additional scrutiny regarding the functionality of the second Mg ion (MgB)⁹⁶.

A comprehensive exploration of the catalytic mechanism of yeast RNAPII by computational means has been carried out by Maria Ramos and colleagues⁹⁷. In order to arrive at an unbiased reaction mechanism Ramos et al. interrogated four independent hypothetical scenarios, all variations on the general two-metal catalysis⁵², but each having distinct yet plausible candidates for proton donor and acceptor. What sets this work apart from other computational interrogations of RNAPII mechanism^84,85 is the combination of the full-atom Gromacs-based MD simulations with thermodynamic integration (TI) free energy calculations, DFT, and hybrid quantum mechanics/molecular mechanics (QM/MM) methods (DFT:PM3MM). This comprehensive combinatorial approach allowed to create a higher quality starting full atom model (before equilibration the fragments missing in the 3.95 Å structure of NTP-bound EC (2e2h) were rebuilt using 2nvz and 2e2i (all⁵⁷)), to carry out high level DFT and lower level PM3MM interrogation of catalytic mechanisms on an expanded (relative to earlier studies^84,85) set of reactants (GTP substrate, 3′-terminal RNA nt, the catalytic tetrad (Asp481, 483, 485, and 837), MgA and B, His1085, Arg446, 766, 1020, and Lys967), and to calculate free energy activation barriers associated with each mechanism⁹⁷. In retrospect, the recently illuminated effect of Watson-Crick base pairing on the k_cat of NTP condensation⁸¹ argues for the importance of incorporation of the templating DNA nt into DFT:PM3MM simulations, however, this was implicitly addressed in the full-atom MD simulation.

Alternate hypothetical mechanisms explored by Ramos and colleagues included the following: 1) acid-base catalysis by the substrate Oα; 2) free OH⁻ hypothesis, 3) MgA-bound OH⁻ hypothesis, and 4) Asp485 acting as general base hypothesis⁹⁷. All these used a model of 3′ OH weakly coordinated by MgA, as the preliminary DFT simulations showed that strongly coordinated MgA-assisted deprotonation of the 3′ OH led to the non-stationary solutions approximately 17 kcal/mol higher in energy, otherwise the strongly coordinated 3′ OH spontaneously converted into weakly coordinated upon movement towards P_α during the nucleophilic attack⁹⁷.

The first mechanism involved the deprotonation of the 3′ OH by the O_α of the substrate (substrate-assisted catalysis), acting first as the base, followed by the nucleophilic attack on the Pα and protonation of the PP_i by the O_α, acting as the acid. Attractive in its simplicity and conforming to the two-proton transfer hypothesis⁷⁴, this reaction mechanism by the way of TI yielded a very high activation barrier for the deprotonation step, resulting in the energy barrier between the reactants and the transition state equal to 34.9 kcal/mol (from the bulk turnover data the authors estimated that the highest energy barrier for NTP condensation is ≤ 18.1 kcal/mol)⁹⁷. Based on TI calculations this mechanism was rejected.

The second reaction mechanism employed free OH⁻ from the bulk solvent as the proton acceptor for deprotonation of the 3′ OH. In estimation by Florian et al., the free energy needed for the OH⁻-assisted deprotonation of the 3′ OH at the neutral pH was 1.5 – 3 kcal/mol, making hydroxide ion from the bulk solvent an attractive replacement of the substrate acting as a general base⁹⁸. This is even lower than Zhu and Salahub estimate for the energy barrier in deprotonation of the 3′ OH by bulk water (7 kcal/mol)⁸⁴. The reaction mechanism according to this scenario comprised transfer of OH⁻ from bulk solvent to the RNAP active site, deprotonation of the 3′ OH by the hydroxide, protonation of the O_β by His1085 acting as general acid, followed by nucleophilic attack on P_α, and protonated PP_i leaving the active site (Fig. 4)⁹⁷. Entry of hydroxide into the active site at neutral pH appeared to be associated with an energy barrier of 7.5 kcal/mol, the highest energy barrier for the entire reaction mechanism reaching 9.9 kcal/mol – about half of the estimate for the barrier of the rate-limiting step in NTP condensation, and the total free energy change of −5.8 kcal/mol, making this reaction pathway feasible both kinetically and thermodynamically⁹⁷.

Figure 4. Two-metal catalysis of NTP condensation by yeast RNA polymerase II.

A. RNA 3′ OH is deprotonated by the OH⁻ from the bulk solvent as substrate’s O_β is protonated by His1085. B. Nucleophilic attack by RNA 3′ O⁻ on substrate’s P_α. C. Substrate in a form of NMP added to the 3′ end of RNA, pyrophosphate in a form of protonated (Mg-PP_i)⁻ leaves active site. This figure was adapted from Carvalho et al⁹⁷, and shows reaction mechanism as a sequence of distinct steps, for clarity and ease of analysis. Proton-inventory experiments reported by Castro et al. indicated that these reaction can take place within the same associative-like transition state⁸².

The next, third reaction pathway explored by Ramos and colleagues, was similar to the second hypothesis, and included the protonation of the leaving PP_i by the His1085, but the deprotonation of the 3′ OH was carried out by MgA-bound OH⁻, instead of 3′ OH-bound hydroxide transferred from the bulk solvent. Positioning the OH⁻ into the coordination sphere of MgA led to significant rearrangements of the reactants, in particular O_2β left Mg²⁺ coordination sphere, whereas pK_a of the O_1β increased to the point of spontaneous abstraction of proton from His1085 and MgA coordination geometry switching to tetrahedral. Overall energy of the Mg-coordinated OH⁻ system was 10.8 kcal/mol higher than the same of the system in second hypothesis containing hydroxide unbound from Mg²⁺. The potential energy surface for the entire reaction shows an energy barrier of 31.7 kcal/mol and the total free energy change of −5.8 kcal/mol⁹⁷. Being thermodynamically identical to the previous reaction mechanism, this pathway is clearly kinetically disadvantageous, yielding an energy barrier more than 3 times of that for free OH⁻ reaction (pathway 2).

The fourth and the final reaction mechanism investigated by Ramos and colleagues was similar to the second scenario modeled in ReaxFF simulations by Zhu and Salahub⁸⁴, where deprotonation of the 3′ OH by catalytic residue Asp485 occurs simultaneously with the nucleophilic attack by the 3′ O⁻ on P_α. The activation free energy for this reaction amounted to 26.2 kcal/mol, lower than in mechanisms 1 and 3, but still higher than that of pathway 2 (9.9 kcal/mol) and the estimate for the NTP condensation rate-limiting step (18.1 kcal/mol); the overall free energy change for the reaction was thermodynamically unfavorable at 14.5 kcal/mol⁹⁷.

The preceding analysis highlights the operational advantage of combining MD simulations with TI, which together allow for more comprehensive ranking of both kinetically and thermodynamically unfavorable reaction pathways. It bears noting that although the NTP condensation along the pathway featuring free OH⁻ from bulk solvent acting as the base, and the His1085 as the general acid emerges as the best candidate, both kinetically and thermodynamically⁹⁷, alternate, kinetically unfavorable mechanisms may govern the action of “slow” variants of RNAP⁴⁴, and/or slow nucleotide addition to paused ECs⁹⁹.

Pyrophosphate release and translocation

The detailed simulations of the NTP condensation and formation of the PP_i reviewed above did not consider the mechanism of PP_i diffusion from the active site post-chemistry. It is, however, an important mechanistic question, as release of PPi not only had been linked to translocation in the so-called “power-stroke” hypothesis (largely based on the studies of the single-subunit RNAP)¹⁰⁰, but is also essential for readying the active site for loading of the next substrate NTP. Multi-subunit RNAPs, including yeast RNAPII, are believed to operate not by the power-stroke mechanism, but to behave instead as isothermal “Brownian ratchet” motors, translocating back and forth along the template with input only of thermal energy^{32,43,44,101,102}. According to this hypothesis PPi release is uncoupled from translocation.

Although consistent with the bulk biochemical, crystallographic, and biophysical data, the isothermal ratchet hypothesis does not provide for an atomic level mechanism of PP_i release, nor the order of translocation and release steps. These were investigated by Huang and colleagues by the means of MD¹⁰³. To this end they had rebuilt missing polypeptides of the 3.95 Å structure of the yeast RNAPII elongation complex (2e2h⁵⁷) using SWISS-MODEL homology modeling software, and restored the 3′ OH omitted in the GTP-bound complex using 1i6h⁶⁷ as the template. Starting model for PP_i release was generated by breaking P_α-O bond in the GTP in 2e2h. MD simulations were carried out using GROMACS 4.5 software with AMBER03 force field. In order to investigate long scale dynamics from short MD simulations the authors employed Markov State Model to identify macrostates using MSMBuilder software.

Huang and colleagues in the final analysis considered the leaving group as unprotonated (Mg-PP_i)²⁻, believing this to be a more prevalent form at physiological conditions and indicating that in the preliminary simulations differently protonated forms behaved similarly¹⁰³. The release of the PP_i was manually driven (steered) in different directions from the active site towards the pore/secondary channel to simulate potential random diffusion routes.

An immediately important observation made by comparing the pre- and post-chemistry simulations, was that the formation of the PP_i lead to significant destabilization of the closed TL. If in pre-chemistry EC fluctuations of the TL tip residues (Gln1078 to Ala1087) were smaller than those averaged over the length of Rpb1 (RMSF of 0.78 Å vs. 1.63 Å), in post-chemistry complex amplitude of the fluctuations of the same residues on average doubled, whereas it remained nearly the same for the reference set¹⁰³. Notably, formation of the PP_i and destabilization of the TL led to loss of interactions between the residue Gln1078 and the substrate 3′ OH, between ribo-discriminating Asn479 and 2′ OH of the substrate, and to the increase in mobility of His1085 and Phe1086. In turn PP_i was observed to establish new contacts, importantly, with Rpb1 Lys752, which together with His1085 formed the initial gateway for PP_i release (Fig. 5). Breaking of the closed TL contacts with the RNA and the increased destabilization of its tip prompted by PP_i formation likely serves as the starting point for conversion of TL into an open configuration and for translocation. Huang and colleagues did stress that they did not observe translocation in their simulation and that in their estimates the time scale of the PP_i elimination from the active site (~1.5 μs) is several orders of magnitude shorter than that of NTP condensation, arguing against coupling of PP_i release and translocation¹⁰³.

Figure 5. Hopping pathway of pyrophosphate release from the active site of yeast RNA polymerase II.

Elements of the active site (from GTP-bound 2e2h) are represented as in Fig. 2. Amino acid residues implicated in facilitating pyrophosphate release are green sticks with green semitransparent spheres.

The simulations of PP_i release did indicate a strong coupling between it and the destabilization of the closed TL: not only the formation of the PP_i led to an increased mobility of the TL tip residues, but in turn the intrinsic motions of the TL appeared to promote the transition from the active site to the second metastable state (“hopping” position S2, Fig. 5) towards the pore/secondary channel¹⁰³. The overall trajectory of the PP_i diffusion was determined to take a form of hopping between four metastable states, sequentially further removed from the active site (Fig. 5). The gateway state S1 forms post-chemistry, where (Mg-PP_i)²⁻ establishes interactions with Lys752 in addition to the pre-existing interaction with His1085. The second state is defined by moving (Mg-PP_i)²⁻ further away from the active site, and adding interactions with Lys619, while maintaining the ones with Lys572 and His1085. These are broken during transition to the third state (S3) where (Mg-PP_i)²⁻ interacts with Lys619 and Lys518. In the next and last state before release of (Mg-PP_i)²⁻ from the pore, it is bound to Lys620 and Lys880. Hopping as the means of facilitated diffusion across long distances is broadly employed in DNA/RNA enzymes, including diffusion of the RNAP itself^104,105.

These simulations also highlighted the stimulatory role His1085 plays in PP_i release, in addition to its role in NTP binding⁷⁹ and in chemistry step (as the general acid^74,97): in the system where this His residue was substituted by Phe, PP_i tended to be trapped in the active site¹⁰³. Such substitutions can and are being explored genetically and biochemically: Kornberg and colleagues reported the lethality of His1085Phe substitution, and the severe defect in in vitro transcription by the His1085Tyr-substituted RNAPII⁶⁸. While these experiments emphasized the functional importance of the targeted residue, amino acid substitutions can and do affect more than one parameter, which can influence the outcome of genetic and biochemical assays in addition to catalysis: protein structure, stability, dynamics (allostery), etc. Alchemical manipulations afforded by MD and QM/MM simulations (such as replacement of His by Phe in post-chemistry model) allow exploration of the effects of such substitutions on each isolated step (e.g. PP_i release) separately, without affecting others (such as chemistry), hence they can not only complement but also aid in interpretation of complex behaviors, observed in vitro and in vivo.

The interplay between translocation, trigger loop dynamics, and the PP_i release was recently explored biochemically by Belogurov and colleagues¹⁰⁶. In an elegant experimental set-up, which allowed them to monitor translocation of E. coli RNAP in real time using quenching of fluorophores incorporated into template DNA, combined with the improved fluorescent detection of free PP_i, the authors were able to demonstrate that translocation and the release of PP_i into solution occur on a similar time scale (t_1/2 of 7–12 ms and 6–9 ms, respectively), the former preceding or occurring simultaneously with the latter¹⁰⁶. Both steps were delayed relative to the chemistry step. Belogurov and colleagues also demonstrated that acting as a Brownian ratchet, RNAP can nevertheless be biased in the direction of translocation through the opening or closing of the active site by the TL, with active site opening favoring forward translocation¹⁰⁶. Taken together with the findings of Da et al.¹⁰³ linking release of PP_i with opening of the TL, one can find features of both the power-stroke¹⁰⁰ and the isothermal ratchet^44,102 in the post-chemistry action of RNAP.

Concluding remarks

Although it might appear that “the basic mechanism of mRNA synthesis during gene transcription is known” from the structural point of view¹⁰⁷, from the studies reviewed herein one can see that the same crystallographic models of RNAPII can be consistent with a number of dramatically different reaction mechanisms, which then have to be ranked kinetically and thermodynamically using complementary methods, such as MD and TI⁹⁷. In fact, many transcription “mechanisms” proposed to date amount to little more than nominative narratives, enumerating the actual or the conjectural steps of a pathway with no regard to physical parameters (such as potential energy surface, activation barrier, initial velocity for molecular collisions, etc.) which would determine their feasibility. Both the structural and the computational explorations of the basic mechanism of transcription illuminate the complexity of this process, most elements of which still elude full realization as an atomic-resolution physico-chemical model. The details of the nucleotide condensation mechanism and its linkage to TL dynamics are starting to emerge and congeal into a physically tractable mechanism. Basic mechanistic and structural foundations for the large scale dynamics of translocation, back-tracking, arrest, initiation, and termination still await their full elucidation.

Highlights.

• NTP condensation by RNA polymerase II occurs via modified two-metal catalysis

• Post-chemistry pyrophosphate release takes place in the enzyme pore via hopping mechanism

• Pyrophosphate release is linked to opening of the trigger loop